Differential Evolution Algorithm for Single Objective Bound-Constrained Optimization: Algorithm j2020
Janez Brest, Mirjam Sepesy Maučec, Borko Boškovič
Abstract
In this paper, a new algorithm is presented to deal with real parameter single-objective optimization problems, which are often complex and computationally very expensive. The proposed algorithm (j2020) is based on the self-adaptive differential evolution algorithms jDE and jDE100. Our algorithm uses two populations like jDE100, while jDE uses only one population. It uses a crowding mechanism, which is not being used in previous algorithms, and a mechanism to choose vectors in the mutation operation from both subpopulations. We provide the obtained results for each benchmark function for four dimension scenarios as required by the organizers of the special session for Single Objective Bound-Constrained optimization. We also compare the obtained results with the original DE and jSO algorithms on the largest dimension scenario.